The quality assessment of water is a challenging task due to extensive experimental requirements. However, the process of water quality monitoring can be automated with the help of Internet of Things (IoT) devices using sensor probes. This article presents the IoT infrastructure-based river water quality monitoring and assessment. Experiments were performed to assess the water quality during different months and seasons for the Ganga River and Sangam (confluence of Ganga and Yamuna rivers) at Prayagraj, Uttar Pradesh, India. The data samples were collected for 15 months continuously using the Libelium smart water kit. The smart water IoT (SWIoT) kit was equipped with sensors to assess specific parameters like pH, dissolved oxygen, temperature, conductivity, and oxidation–reduction potential. An algorithm is also presented that harnesses principal component analysis and factor analysis for feature selection and weight assignment for river water quality assessment. Further water quality is quantified using the water quality index that helps to categorize the water quality for different usages. The results corroborate that the water quality of the Ganga River was found to be better than the Sangam site most of the time, owing to the higher level of pollution in Yamuna River. Additionally, the water quality of both rivers was found to be suitable for irrigation and fisheries but not for drinking purposes, considering the average oxygen levels.

Compressed sensing based data gathering is in-efficient in small scale wireless sensor networks because of uncorrelated observations at the sink. Failure to exploit the low rank and suitable transform domain to explore the correlation structure of sensor data, results in significantly low recovery accuracy in such matrix completion algorithms for small scale networks. Targeting the spatio temporal correlation structure of the data, a novel data gathering and matrix completion scheme, which exploits the low rank property and compactness of sensor data which exists in wavelet transform domain, is proposed in this work. The compactness of the sensor data in wavelet transform domain is used for recovering the missing entries of the matrix. Experiments and simulations, for single-node and multi-node scenarios, prove the efficacy of the proposed approach over existing schemes significantly in terms of recovery accuracy even at extremely low sampling rate.